A new method by Intel is being developed, which can dramatically lessen the amount of gaming textures, saving VRAM and storage space. This method, which functions similarly to Nvidia’s neural texture compression, is intended to reduce file size while maintaining image quality.
According to Intel, their method may reduce textures by up to 18 times in a more aggressive mode and up to 9 times in quality mode. Additionally, there will be two iterations of the technology.
Whilst the other can run on standard CPUs as well as GPUs but can be slower, the previous is made for Intel’s XMX hardware for improved performance. The method reduces texture sizes by combining sophisticated math with existing compression (BC1).
It employs trained data to maintain textures’ same appearance even after they are shrunk in size, as opposed to just compressing photos as usual. The textures are compressed by an encoder and then restored when necessary by a decoder. A simpler approach (FMA), which is slower but requires more hardware, is used in a fallback version.
According to Intel, there are four primary methods that developers can utilise this technology. Compressing textures before downloading is one way to reduce installation time and storage usage for games. While playing, the other strategies are effective. Textures can be loaded as soon as the game launches, streamed throughout gameplay, or loaded just when necessary without being stored in VRAM.
Systems with little memory can benefit from this final option. There are two modes in the system: variation A and variant B. Variant A emphasises maintaining high standards of quality. For instance, while maintaining the same quality, big 4K textures can be shrunk from 64MB to 10.7MB. The size and quality of other textures are somewhat diminished.
Variant B applies more forceful compression. While certain textures are further compressed and have their resolution lowered to as low as 0.17MB, others are kept at full fidelity. Intel tested its solution against more antiquated compression techniques.
While the earlier approach only achieved roughly 4.8x compression, variants A and B obtained about 9x and 18x, respectively. As a result, Intel’s performance is comparable to Nvidia’s strategy, particularly in the more aggressive mode.
Compared to certain rival technologies, Intel claimed that its solution is more compatible with a larger variety of hardware, including non-Intel graphics cards.
